AMD GPU Driver (amdgpu) 31.30.0 release notes

AMD GPU Driver (amdgpu) 31.30.0 release notes#

The release notes provide release highlights and resolved issues since the previous AMD GPU Driver release (31.20.0, a preview release).

Release highlights#

The following are notable new features and improvements in AMD GPU Driver 31.30.0.

For compatibility between AMD GPU Driver, ROCm, GPUs, and operating systems, see the Compatibility matrix.

Broader hardware enablement#

  • New SMU power-management generation v15.0.8 with full clock, thermal, power-limit, and metrics reporting on next-generation AMD platforms.

  • New system-DMA (LSDMA v7.1) and PSP 13.0.15 support for upcoming products.

  • APU + dGPU (A+A) platform enablement on GMC 12.x with correct VRAM sizing, aperture layout, and page tables for unified-memory systems.

Compute and AI workload improvements#

Hardened user-mode queues (userq) with stronger validation, safer doorbell handling, and cleaner create/teardown to reduce hang and resource-leak risk under heavy AI/ML submission.

Power, thermal, and telemetry#

  • Per-component temperature reporting (AID, XCD, HBM) plus baseboard and GPU-board metrics for data-center monitoring.

  • User-tunable GFX clock ranges, custom FCLK (OD_FCLK), and zero-fan OD on SMU v13/v14 for finer control over performance and acoustics.

  • NPM (Node Power Management) on SMU v15.0.8 for platform-level power budgeting.

Stability and security#

  • Broad NULL-pointer, use-after-free, and race-condition hardening in device init/teardown, VM acquisition, PASID reuse, and RAS paths.

  • Bounds checking on MMHUB/GMC client-ID decoders to prevent out-of-range reads when decoding page-fault sources.

Resolved issues#

The following previously known issues have been resolved in this release:

  • Resolved KFD/ROCm workload startup failures on hosts with non-4K page sizes.

  • Resolved a GPU idle power consumption regression on the latest GFX v12 discrete GPUs.

Contents